Dispatch R589

Color vision: Putting it togetherPeter Lennie

Color vision depends on the visual system comparingsignals that originate in different classes of conephotoreceptors. New work shows that the differentclasses of cones are not only distributed irregularly, butin different individuals they are present in very variableproportions. Surprisingly, this does not affect color vision.

Address: Center for Neural Science, New York University,4 Washington Place, New York, New York 10003, USA.E-mail: peter.lennie@nyu.edu

Current Biology 2000, 10:R589–R591

0960-9822/00/$ – see front matter © 2000 Elsevier Science Ltd. All rights reserved.

The fundamentals of color-vision have been well-understood for a long time. By the early 19th century,Young had deduced that the normal human eye containedthree sensing mechanisms selectively tuned to differentbut overlapping regions of the visible spectrum. Each ofthese mechanisms — now known to be distinct classes ofcone photoreceptors — signals the number of photons itabsorbs, but nothing about the spectral composition of thelight. Changing the spectral composition of the light altersonly the probability that photons will be absorbed by acone. The color signature of any point in a scene is there-fore represented in the nervous system by just threenumbers that specify the photon catches in each of thethree classes of cones. This is known as trichromacy.Many physically different spectral distributions of lightcan give rise to the same three numbers, and the normalvisual system will see these as identical. This fact is thefoundation of color rendering systems that create a widerange of colors by mixing a small number of primarysources in appropriate quantities.

The principle of trichromacy tells us which differentspectral mixtures will look identical, but it does not tellus why these mixtures look the way they do. Interestingobservations about color appearance that are beyond thereach of the principle of trichromacy led Hering tosuggest in the mid 19th century that the mechanisms ofcolor vision are organized as three pairs of polar oppo-sites: light–dark, red–green and yellow–blue. The princi-ple of opponent organization explained an array offacts about color appearance — including why red,green, yellow and blue have special stature as anchoringcolors; why one does not experience reddish–greenhues; and why particular mixtures appear white. But thethree mechanisms on which color opponency was sug-gested to depend are not like those postulated toaccount for trichromacy.

Trichromacy and color-opponency are reconciled byrecognizing that they characterize different stages in theanalysis of color: the three classes of cones constitute thefirst stage, and their signals are later combined at a secondstage that gives rise to opponent mechanisms. Theseprinciples, which have been established through studies ofhuman perception, provide sharp guidance to biologistsstudying the underlying machinery, and have madephysiological studies much more incisive than they couldhave been otherwise. The big ideas have been reassur-ingly confirmed over the past thirty years [1,2], but recentstudies show unsuspected and intriguing complexities inthe organization of chromatic mechanisms in the retinaand at later stages of the visual pathway.

One puzzle is highlighted in a collection of paperspublished recently in the Journal of the Optical Society ofAmerica on the “chromatic topography of the retina” [3]. Tounderstand how the chromatic machinery is put together,we need to know the proportions and distributions of thelong-wavelength sensitive (L), middle-wavelength sensitive(M) and short-wavelength sensitive (S) cones. The S conesare morphologically and histochemically distinctive; theyconstitute about 8% of cones and are arranged in a quasi-crystalline mosaic [4]. Until very recently, the L and Mcones were enigmatic. Individual cones could be identifiedas L or M by measuring their spectral characteristics, butwe had no secure means of assessing their relative numbersor distributions on the retina. The difficulty probablystems from the fact that distinct L and M cones — reliablypresent only in old-world primates — are a recent evolu-tionary development from a single ancestral cone type [5],and their genes and gene products are very much alike.

Recent work has substantially cracked this problem. Thenew papers assemble converging evidence — fromperceptual experiments, analysis of the amounts of L andM cone mRNAs in the retina, very high-resolution imagingof the photoreceptor mosaic in the living eye, and record-ing of the electroretinogram — that L cones outnumber Mcones by about 1.6:1 in the average eye. Much moreintriguing is the finding that this ratio varies by at least afactor of four among individuals, yet this seems to havenegligible effect on their color vision [6]. We would notexpect variations in cone ratios to affect the underpinningsof trichromacy, because physically different mixtures oflights that cannot be distinguished give rise to identicalsignals in every cone, but we might be very surprised thatdifferent ratios of cones have so little effect on the oppo-nent mechanisms that gather and compare signals from dif-ferent classes of cones, and do influence appearance.

There is an additional puzzle: using an adaptive opticalsystem tuned to compensate for local aberrations in theoptics of an individual’s eye, Roorda and Williams [7] haveobtained astonishingly high-resolution images of themosaics of L, M and S cones in the normal living eye, andhave shown not only that there are large inter-individualvariations in the L:M cone ratio, but that neither mosaic is atall regular, instead being clumpy, possibly random. All thisought to frustrate the assembly of the red–green opponentmechanisms that compare L and M cone signals locally inthe image, and variable color perception should result.

We know that opponent mechanisms arise in the retina:midget ganglion cells, whose axons convey visual infor-mation from the retina to the brain, are color-opponent,being excited by light in one wavelength band and

inhibited by light in another. Each cell picks up signalsfrom cones in a small region of retina — its ‘receptivefield’, organized in concentric ‘center’ and ‘surround’regions that antagonize each other, one being excited byillumination, the other inhibited (Figure 1). We knowthat, in most receptive fields, center and surround havedifferent spectral sensitivities [8], and it has been gener-ally presumed that center and surround are each fedsignals from a single class of cone. But given wide varia-tion across individuals in the relative proportions of theL and M cones, and irregular arrangements of them, howcan one assemble a small receptive field that segregatesdifferent classes of cones in center and surround?

An intriguing answer to this question, for which evidenceis accumulating, is that the visual system does not botherto, or cannot, do this in any principled way. Instead itrelies on a simple expedient: each receptive field forms itscenter from one or a few juxtaposed L and M cones, andforms its surround by drawing indiscriminately on all Land M cones in a larger surrounding area. If the centerdraws input entirely or mostly from a single cone — verylikely to happen in the fovea and nearby retinal regions —and not all cones in the surround are of the same class,then the neuron must be color-opponent. When receptivefield centers become larger, as they do with increasingdistance from the fovea, the center will draw on severalcones, which might not all be of the same class. Color-opponency will therefore be weakened, leading to a corre-sponding perceptual decline, but even in the retinalperiphery the clumping of cones will ensure that somecenters gather signals from cones of a single class.

Until now there has been no direct evidence for this idea,but it draws support from a series of important findings byDacey and colleagues [9–11], who have recorded signalsfrom generally inaccessible neurons in the primate retina. Land M cones provide signals to both centers and surroundsof midget ganglion cells in peripheral retina [9]. Bipolarcells, the intermediaries between cones and ganglion cells,have center–surround receptive fields like those of gan-glion cells, and probably account for the ganglion cells’receptive fields [10]. The bipolar cell’s surround seems toarise in the H1 class of horizontal cell, which collects signalsfrom only L and M cones in proportions that vary greatlyfrom cell-to-cell and probably reflect the numbers of L andM cones available locally in the retina [11].

Retinal circuitry thus seems to be haphazard in its dealingswith L and M cones, yet its indiscriminate handling oftheir signals finds no expression in our color vision. Howdoes the visual system mask the huge variation, within andbetween individuals, in the signals L and M cones provideto chromatic mechanisms? Part of the answer probably liesin the way that ganglion cells, through bipolar cells, weightheir connections with cones in the receptive field’s center

R590 Current Biology Vol 10 No 16

Figure 1

How red–green color-opponent mechanisms might be formed in theretina. The different classes of cone photoreceptors are distributedirregularly in the retinal mosaic, illustrated at the bottom of the diagram.The short-wavelength sensitive (S) cones, colored blue, constitute lessthan 10% of the total, and are not part of the red–green mechanism.The middle-wavelength sensitive (M) cones, colored green, and thelong-wavelength sensitive (L) ones, colored red, are present inproportions that vary widely among individuals, with the L conesoutnumbering the M cones, on average, by perhaps 1.6:1. The red–green opponent mechanism, two examples of which are depicted bythe smaller circular arrays at the top of the diagram, probably has itsorigin in the midget bipolar cell, which is connected directly to a singlecone that forms the center of its receptive field, and indirectly, viahorizontal cells, to a pool of neighboring cones that form theantagonistic surround. The small circles in the circular arrays eachrepresent a cone in the receptive field, with color representing theweight of the signal (hence the central cone is in a saturated color, andthose in the surround are paler, representing their lower weight). Thestrength of color-opponency depends on the particular pool of signalsthat make up the surround. In the two examples shown here, the singleL cone in the center, identified by the line connecting it to the mosaic,is opposed in one case by a surround that contains few M cones(leading to weak color-opponency), and in the other case by a surroundthat contains predominantly M cones (leading to strong opponency).

Current Biology

Color-opponent receptive fields

Mosaic of retinal cones

and surround. We know that, in general, center and sur-round are equal in their aggregate sensitivity — that is,when one illuminates the receptive field uniformly, thecell does not respond. If this normalization is an organiza-tional imperative, then variations in the proportions of Land M cones in the retina will be substantially discountedautomatically [12]. More stability in color perception iscontributed by the way in which later stages of the visualpathway aggregate the signals from color-opponent cells inthe retina. Lots of perceptual evidence points to the factthat we sense color variations in images on a relativelycoarse spatial scale — coarser than the sizes of retinalreceptive fields. Mechanisms in the cortex must thereforecollect signals from many retinal inputs, thereby smooth-ing-out the local variations among them.

References1. De Valois RL, Abramov I, Jacobs GH: Analysis of response patterns

of LGN cells. J Opt Soc Am 1966, 56:966-977.2. Derrington AM, Krauskopf J, Lennie P: Chromatic mechanisms in

lateral geniculate nucleus of macaque. J Physiol (Lond) 1984,357:241-265.

3. Roorda A, Neitz J: Chromatic topography of the retina. J Opt SocAm A 2000, 17:495-650.

4. de Monasterio FM, Schein SJ, McCrane EP: Staining of blue-sensitive cones of the macaque retina by a fluorescent dye.Science 1981, 213:1278-1281.

5. Mollon JD: Uses and evolutionary origins of primate colour vision.In Evolution of the Eye and Visual System. Edited by Cronly-Dillon J,Gregory RL. Macmillan: Basingstoke; 1991:306-319.

6. Brainard DH, Roorda A, Yamauchi Y, Calderone JB, Metha AB,Neitz M, Neitz J, Williams DR, Jacobs GH: Functional consequencesof the relative numbers of L and M cones. J Opt Soc Am A 2000,17:607-614.

7. Roorda A, Williams DR: The arrangement of the three cone classesin the living human eye. Nature 1999, 397:520-522.

8. Wiesel TN, Hubel DH: Spatial and chromatic interactions in thelateral geniculate body of the rhesus monkey. J Neurophysiol1966, 29:1115-1156.

9. Dacey DM: Circuitry for color coding in the primate retina.Proc Natl Acad Sci USA 1996, 93:582-588.

10. Dacey D, Packer OS, Diller L, Brainard D, Peterson B, Lee B: Centersurround receptive field structure of cone bipolar cells in primateretina. Vision Res 2000, 40:1801-1811.

11. Dacey DM, Diller LC, Verweij J, Williams DR: Physiology of L- andM-cone inputs to H1 horizontal cells in the primate retina. J OptSoc Am A 2000, 17:589-596.

12. Lennie P, Haake PW, Williams DR: The design of chromaticallyopponent receptive fields. In Computational Models of VisualProcessing. Edited by Landy MS, Movshon JA. MIT Press: Cambridge,MA; 1991:71-82.

Dispatch R591